APBRmetrics

http://bkref.com/tiny/5yCkJ 8 lineups used over 150 minutes are over +5 per 100 possessions. Only Blazers are pushing that lineup over 20 minutes per game. How many, if any of the other top performers will do so by season end or especially in playoffs?

4 below -5. How many of the bottom 3 will cut back minutes or eliminate that lineup altogether? Might be hard to do worse than that. Or explain it as bad luck / learning?

For Charlotte top 5 lineups are horrible, worse than team average and way worse than team average without those lineups. http://www.basketball-reference.com/tea ... 5/lineups/ Bad coaching / analysis or just early, give it time? Last season, only 1 in top7 was much above neutral and they did not retain McRoberts who was in it.

Magic, only one in top 7 positive. Starers with Ridnour, not much touted Oladipo or Payton. http://www.basketball-reference.com/tea ... 5/lineups/
Might be mostly against subs but still... huge difference. Worth trying the good performing lineup more than 2 minutes a game? Ya think? Last season only 1 of top 6 lineups was above neutral, barely. Lots of personnel change but not in top lineup performance.

C. Boozer | K. Bryant | J. Hill | W. Johnson | J. Lin -20.6. http://www.basketball-reference.com/tea ... 5/lineups/ Played almost as much as next 19 lineups. You can overdo lineup concentration, but compared to this, it might be hard. About one third of the lighter used 19 were positive.

Crow,

I'm gonna be getting into a project on lineup analysis in the NBA beginning next week and Ill be working on it for the next few months. I hope I find some interesting stuff, and if/when I do Ill try to update you guys here.

Alright, thanks.

If you find a good, revealing way to cluster lineups based on player composition by position with less than 240 or 7000 clusters let me know. Maybe doing it without player composition or position orientation is the way to go, just lineup stats, maybe by four factors? Still can you get it much below 80 clusters, down to maybe 20 fairly similar super clusters? You could do a lot of further work on 20 lineup clusters including matchup analysis.

You read my mind. I think that grouping lineups by some type will be the trickiest part.

The number of clusters I mentioned above would come from characterizing each position into just 3 types or tiers overall or on offense and defense. That gets too much for easy analysis. Maybe for a team it would still be worthwhile to analyze that world or an even more complex lifelike one. But stepping back from that it is probably looking at lineups as working units. David Sparks analyzed lineups by 7 player types. That yielded some insights of lineup distributions and matchup results. But you probably need at least 7 player types and 3 quality levels still to model things semi-accurately that way and distinguish between good, average and poor lineups with surface similarity.

2 things the 25 best lineups used over 50 minutes have on average compared to opponents? More 3pt attempts and FT attempts.